HAL Id: hal-01533580
https://hal.inria.fr/hal-01533580
Submitted on 6 Jun 2017
HAL
is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire
HAL, estdestinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Distributed under a Creative Commons
Attribution| 4.0 International LicenseEffective Packet Loss Estimation on VoIP Jitter Buffer
Miroslav Voznak, Adrian Kovac, Michal Halas
To cite this version:
Miroslav Voznak, Adrian Kovac, Michal Halas. Effective Packet Loss Estimation on VoIP Jitter Buffer. Networking Workshops (NETWORKING), May 2012, Prague, Czech Republic. pp.157-162,
�10.1007/978-3-642-30039-4_21�. �hal-01533580�
Effective Packet Loss Estimation on VoIP Jitter Buffer
Miroslav Voznak1, Adrian Kovac2, Michal Halas2
1 VSB – Technical University of Ostrava, 17. listopadu 15, 708 33 Ostrava-Poruba, Czech Republic
miroslav.voznak@vsb.cz
2 STU – Slovak University of Technology, Faculty of Electrical Engineering, Ilkovicova 3, 812 19 Bratislava, Slovak republic
{kovaca, halas}@ktl.elf.stuba.sk
Abstract. The paper deals with an influence of network jitter on effective pack- et loss in dejitter buffer. We analyze behavior of jitter buffers with and without packet reordering capability and quantify the additional packet loss caused by packets dropped in buffer on top of the measured network packet loss. We pro- pose substitution of packet loss parameter Ppl in ITU-T E-Model by effective packet loss Pplef incorporating network jitter, a jitter buffer size and a packet size as additional input parameters for E-Model.
Keywords: Jitter buffer; Dejitter buffer; MOS; E-Model; Packet loss; Packet drop; Pareto distribution; Pareto/D/1/K; VoIP.
1 Introduction
Voice over IP communication gains still greater importance in telecommunications industry. Lack of synchronization in comparison to TDM (Time-division Multiplex- ing) brings concerns about variable conditions on network which cause packet loss and fluctuations in delay (jitter). Jitter causes excess packet loss on receiving buffers depending on the buffer size and delay variance. When using E-Model as an objective method based purely on network parameters for speech quality assessment, effects of jitter are not incorporated in original E-Model. Estimated MOS (Mean Opinion Score) can be positively biased or be too optimistic about call quality under real IP network conditions. In our paper we propose methods of numerical approximation of general jitter buffer behavior. These approximations are used to quantify effective network packet loss Pplef taking three additional network parameters on the input: jitter buffer size [ms]; voice packet size [ms] and network delay variance (jitter) [ms].
2 Selected Time-Related IP Network Parameters
Under time-related network parameters we understand packet transmission delay, mean interarrival time difference – jitter and secondarily packet loss, which can be
understood as infinite delay of packet delivery. Based on practical experience of voice perception and human conversational model, ITU-T G.114 [1] defines the recom- mended value of Mouth-to-Ear delay which consists of partial delays occurring at different stages of communication path. The delay distribution on IP networks can be successfully modeled and described by Pareto distribution which give better results compared to Weibull, Poisson or Log-Normal distribution [2]-[4], [11]. Pareto distri- bution belongs to family of geometrical and long-tailed distributions that characterize statistical set where most values are clumped around the beginning of the interval.
Equation (1) shows PDF and equation (2) CDF functions of generalized Pareto distri- bution (GPD).
( )
( ) ( )
1
, , 1 1 x
F x
ξ ξ µ σ
ξ µ
σ
−
−
= − +
( )
( ) ( )
1 1
, ,
1 1 x
f x
ξ ξ µ σ
ξ µ
σ σ
− −
−
= +
(1), (2)
Where σ = scale, ξ = shape and µ = location parameter (min. value of random varia- ble with Pareto distribution), µ is an offset of Pareto curve from zero and represents minimal network delay Ta-min. Jitter J [ms] is calculated in real-time as floating aver- age of 16 samples of differences between interarrival times (timestamps) of consecu- tively received packets contained in RTP (Real Time Protocol) header, the calculation is defined in RFC 1889 and given by equation (3) where each difference is calculated according to equation (4). Jitter value is transferred in RTCP (Real Time Control Protocol) protocol header as one of the QoS parameters.
J = J + (|D(i-1,i)| - J) / 16 [ms] and D(i, j) = (Rj - Ri) – (Sj – Si) [ms] (3), (4) Where R are timestamps of packet reception time, S when packet was sent and indices i,j are consecutive packet numbers.According to G.1020 [5] an alternative approach of jitter based on determination of the Mean Absolute Packet Delay Variation with regard to a short term average or minimum value (adjusted absolute packet delay variation) offers more accurate calculation short-term jitter better describing relation- ship to jitter buffer behavior. In [5] the short term jitter is computed for current packet (i) whose delay is designated ti. Packet (i) is compared to a running average estimate of the mean delay using 16 previous packet delays and assigned either a positive or negative deviation value. These characteristics are calculated according to equations (5), (6) and (7) [5], where a Mean Delay is expressed in relation (5), a Positive dealy deviation in (6) and Negative deviation in (7).
Di = (15 × Di-1 + ti-1) / 16 (5)
Pi = ti – Di if ti > Di and Ni = Di – ti if ti < Di (6), (7) If ti = Di, then both Pi and Ni are zero. Mean Absolute Packet Delay Variation 2 (MAPDV2) for packet i is computed as (8) [5]:
MAPDV2 = mean(Pi) + mean(Ni) (8)
Variance of input stream packet delivery causes problems with synchronous play- back. Receiver can wait finite amount of time for data to be delivered [12], [13]. All data that are not lost in the network but arrive later than expected are considered lost, hence we use the term Effective packet loss to describe all losses in data path. Effec- tive packet loss is encountered at the input of audio decoder and is equal to or greater than measured network packet loss advertised through RTCP packets. We analyze additional loss on jitter buffer and its inclusion into E-Model further.
3 E-Model as Objective Measurement Method
E-Model defined by ITU-T G.107 is widely accepted objective method used for estimation of VoIP call quality [6], [7]. E-Model uses set of selected input parameters to calculate intermediate variable, R-factor, which is mostly converted to MOS-CQ value. Input parameters contribute to the final estimation of quality in additive manner as expressed in (9).
R = Ro – Is – Id – Ie-eff + A . (9) Where Ro represents the basic SNR, circuit and room noise; Is represents all im- pairments related to voice recording such as quantization distortion, low voice volume, compression artifacts; Id covers degradations caused by audio signal de- lay including side tone echo; Ie-eff impairment factor represents all degradations caused by network transmission path, including end-to-end delay, packet loss, codec compression artifacts and PLC (Packet Loss Concealment) masking capa- bilities and A is an advantage factor of particular communication technology. We focus at Ie-eff parameter, which is calculated as in (10):
Ie-eff = Ie + (95 – Ie) × Ppl / (Ppl + Bpl) . (10)
Where Ie represents impairment factor given by codec compression and voice repro- duction capabilities, Bpl is codec robustness describing immunity of particular codec against random losses and quantifying its PLC masking qualities. These values are listed for narrowband codecs in ITU-T G.113 appendix [8]. We propose to substitute Ppl parameter for an overall effective packet loss Pplef according to equations (11) and (12). In this document we refer to “Modified E-Model” as to model based on original ITU-T G.107 with Ppl packet loss substituted by proposed Pplef. Both Ppl and Pplef values lies in interval <0,1>.
Pplef = 1 – (1 – Ppl).(1 – Pjitter) and Pjitter = (Pplef - Ppl)/(1 – Ppl) (11), (12) Where Pjitter is net packet loss on jitter buffer. According to the type of buffer used, Pplef should be substituted by either Ploss_wo or Ploss_wr respectively.
3.1 Jitter Effects Simulation and Measurements of Effective Packet Loss Measurements, which were carried out, proved the jitter buffer behavior described in chapter 2 and helped to find a best fitting function modeling the jitter buffer packet
loss Pjitter. We have simulated MOS dependence of several codecs such as G.711 both µ-law and A-law with PLC, G723.1 (ACELP and MP-MLQ), G.726, G.729 on QoS parameters which were presented by One-way delay Ta ∈{0, 20, 50, 100, 150, 200, 300, 400} [ms]; Network packet loss Ppl ∈{0, 1, 2, 3, 5, 7, 10, 15, 20} [%] and Net- work jitter of 20, 40 and 80 ms with Pareto distribution.VoIP traffic was emulated using IxChariot software and the transport network presented a computer with two network cards with WANem software implementing Pareto distribution on an output packet stream. The stream was received with third computer with IxChariot endpoint, the situation is depicted in Fig. 1. For each codec and combination of QoS parameters values were simulated one-minute VoIP calls and repeated three times, from which data for evaluation were collected, overall 189 combinations of parameters were si- mulated [9], [10].
Fig. 1. Test bench for VoIP MOS evaluation with E-Model.
Based on simulation results and measurements we have determined optimal shape parameter ξ giving the smallest overall MSE error of differences between measured and estimated Ploss_wo and Ploss_wr by equations (13) and (14). Optimal value of sought shape parameter ξ is around values – 0.1 to – 0.2 depending on actual network traffic characteristics, but in general value of – 0.1 proved itself to give good results across wide range of LAN IP networks.
( )
ξσ µ ξ
1
wo _ loss
1 x P
−
−
+
=
( )
2 .1 1 x
P
1
wr _ loss
ξ
σ µ ξ −
−
+
= (13), (14)
After substitution of equations (13) and (14) into (12), parameters ξ = – 0.1 and µ = 0 we get equation for jitter buffer packet loss without reordering (equation (15), x = buffer size in [ms]) and with reordering capability (16) where x = packet size in [ms]:
10 wo
_ loss
x 1 . 1 0
P
−
+
= σ 2
.1 x 1 . 1 0 P
10 wr
_
loss
−
+
= σ (15), (16)
To use these calculated losses in should be used in place of parameter
proposed equation for equipment impairment factor calculation in Pplef = 1 – (1
I Graph in Fig. 2 show implementation of ITU
modifications in case of G.729
additional parameters: network jitter, buff
Proposed modifications give more accurate MOS showing good correlation with real network conditions with jitter as opposed to too optimistic original E
ceived similar results for other measured codecs.
Fig. 2. MOS,
4 Conclusions
We propose the E-Model improvement based on the meter Ppl by the effective packet loss
of this paper. Since the
sider the characteristics constant if they are similar over recent 16 samples. When considering codec with 20 ms audio per packet,
ms. The transient response of real for practical purposes below 1 second Acknowledgement
This work was supported by IT4Innovations Centre of Excellence project, reg. no.
CZ.1.05/1.1.00/02.0070 within Ope
Innovations’ and by project VEGA No. 1/0186/12 „Modeling of Multimedia Traffic Parameters in IMS Networks“
lava.
To use these calculated losses in E-Model we propose to use equation (17 should be used in place of parameter Ppl (network packet loss). Equation (18 proposed equation for equipment impairment factor calculation in E-Model I
(1 – Ppl) × (1 – Pdejitter) = Ppl + Pdejitter – Ppl× Pdejitter
Ie,eff = Ie + (95 – Ie) × Pplef / (Pplef – Bpl)
shows comparison of: measured MOS, MOS estimated by original implementation of ITU-T E-Model and MOS estimate using E-Model with proposed in case of G.729 codec. Modified E-Model takes into account three additional parameters: network jitter, buffer size [ms] and audio packet length [ms].
Proposed modifications give more accurate MOS showing good correlation with real network conditions with jitter as opposed to too optimistic original E-Model
ceived similar results for other measured codecs.
MOS, G.729 codec, network jitter 20 ms, buffer 40 ms
Model improvement based on the substitution of packet loss par effective packet loss Pplef and we perceive it as the main contribution this paper. Since the jitter J is calculated as floating average of 16 packets, we co sider the characteristics constant if they are similar over recent 16 samples. When considering codec with 20 ms audio per packet, so all values are balanced after 320 s. The transient response of real-time E-Model shows sufficiently fast recovery rate for practical purposes below 1 second.
was supported by IT4Innovations Centre of Excellence project, reg. no.
CZ.1.05/1.1.00/02.0070 within Operational Program ’Research and Development for project VEGA No. 1/0186/12 „Modeling of Multimedia Traffic Parameters in IMS Networks“ conducted at Slovak University of Technology Brati
17), which 18) is final
Ie-eff. (17) (18) comparison of: measured MOS, MOS estimated by original Model with proposed Model takes into account three er size [ms] and audio packet length [ms].
Proposed modifications give more accurate MOS showing good correlation with real Model, we re-
substitution of packet loss para- and we perceive it as the main contribution is calculated as floating average of 16 packets, we con- sider the characteristics constant if they are similar over recent 16 samples. When
all values are balanced after 320 sufficiently fast recovery rate
was supported by IT4Innovations Centre of Excellence project, reg. no.
rational Program ’Research and Development for project VEGA No. 1/0186/12 „Modeling of Multimedia Traffic at Slovak University of Technology Bratis-
References
1. International Telecommunications Union, ITU-T Recommendation G.114: One-way Trans- mission Time, ITU-T, Geneva (2003)
2. Koh, Y., Kiseon K.: Loss Probability Behavior of Pareto/M/1/K Queue. In: Communications Letters, vol.7, no.1, pp. 39--41, IEEE Press, New York (2003)
3. Mirtchev, S., Goleva, R.: Evaluation of Pareto/D/1/K Queue by Simulation. In: Information Science and Computing, International Book Series, Technical University of Sofia, Sofia (2008)
4. Dang, T.D., Sonkoly, B., Molnar, S.: Fractal Analysis and Modeling of VoIP Traffic. In:
11th International Telecommunications Network Strategy and Planning Symposium NETWORKS 2004, pp. 123--130, Vienna (2004)
5. Morton, A. et al.: Draft new Recommendation G.1020 (formerly G.IPP): Performance Pa- rameter Definitions for Quality of Speech and other Voiceband Applications Utilizing IP Networks, TIA (2003)
6. International Telecommunications Union, ITU-T Recommendation G.107: The E-Model, a computational model for use in transmission planning, rev. 8, ITU-T, Geneva (2000) 7. Voznak, M. E-model Modification for Case of Cascade Codecs Arrangement , International
Journal of Mathematical Models and Methods in Applied Sciences, Issue 8, Volume 5, (2011)
8. International Telecommunications Union, ITU-T G.113, Appendix I: Provisional Planning Values for the Equipment Impairment Factor Ie and Packet-Loss Robustness Factor Bpl, ITU-T Recommendation G.113, ITU-T, Geneva (1999)
9. Kovac, A., Halas, M., Orgon, M.,Voznak, M.: E-model MOS Estimate Improvement through Jitter Buffer Packet Loss Modelling, In Journal Advances in Electrical and Elec- tronic Engineering , Volume 9, Number 5, (2011)
10. Kovac,A., Halas, M.,Voznak,M.: Impact of Jitter on Speech Quality Estimation in Modified E-model, In conference proceedings RTT 2010, Velke Losiny, Czech Republic, (2010) 11. Mirtchev, S., Statev, S.I.: Study Of Queuing Systems With A Generalized Departure
Process. In: Serdica, Journal of Computing, vol.2, no.1, pp. 57--72, Sofia (2008)
12. Broitman, M., Matantsev, S.: Determination of Optimal Size of the De-jitter Buffer of a Receiving Router. In: Automatic Control and Computer Sciences, 2008, vol.42, no.3, pp.
133—137, Allerton Press, Inc. (2008)
13. International Telecommunications Union, STUDY GROUP 12 – Delayed Contribution 98:
Analysis, Measurement and Modelling of Jitter, ITU-T, Geneva (2003)
